Tag
The paper introduces Verdict-MQ, a 395M open encoder that encodes an input's state once and answers many typed questions about it via a structured attention mask, outperforming the 421M Laya cross-encoder on calibration and accuracy for typed decisions while trading off performance on RACE and BoolQ.
The paper introduces SBERT2S1, which converts Sentence-Transformers retrieval encoders into calibrated typed decision models, along with the BIODECIDE benchmark and MEDLINE-S1 dataset (243k decisions from NLM indexing), showing that retrieval training helps prior-fused residual heads but often hurts cross-heads, and that RLCD reward training underperforms cross-entropy due to biased reward normalization.
This article demonstrates how to adapt the GLM-5.3-Flash LLM to function as a typed decision model similar to Jev, achieving comparable accuracy and speed while enabling decisions on images in a single forward pass.
Bonsai-Llama-Jev is an open-source, vision-enabled typed-decision inference system that runs locally with low VRAM and high accuracy, outperforming other systems in a diverse benchmark.
This article curates demos, tools, skills, and integrations for Jev, a typed decision model from TypeSafe AI, providing a starting point for developers.
AgentJev-0.6B is a small open-source AI decision model that outperforms Laya on Typed Decisions with improved accuracy and significant computation reduction, enabling faster local inference.
Laya-MLX is an open-weight tool for running typed decision AI models locally on Apple Silicon with low latency, providing native inference without cloud APIs. It includes benchmarks showing fast performance on devices like M3 Max.